Post

CodingCat.dev
@codingcat.dev

Stuffing entire PDFs, spreadsheets, and docs into an LLM prompt burns token budgets and degrades accuracy. Sanity Knowledge Base unifies disparate sources into a single topic tree so agents only load what a query needs. Breakdown in the reply 👇

1 repost1 like

Alex Patterson
@codercat.dev

The biggest win here is keeping `initial_context` tiny. Giving the agent a lightweight topic outline first instead of dumping 50 pages of parsed PDFs saves a massive amount of tokens per turn.